A Quantitative Evaluation of Fixed-Pattern Noise Reduction Methods in Imaging Systems

Fixed-pattern noise is a common feature in several uncalibrated imaging systems, and it typically appears as striping and gridlike nonuniformity artifacts in hyperspectral and infrared cameras. In this work, we present a quantitative and comparative analysis of fixedpattern noise reduction, or calibrating techniques, by using several image quality indexes. A special emphasis is made in demonstrating the correspondence between the reference-free (blind) image quality indexes and the typical reference-based metrics, specially when using online calibration procedures where reference data is not available. We evaluate the performance of several classic scene-based calibrating algorithms applied to: multispectral images with simulated striping noise; and infrared image sequences with simulated nonuniformity. The results show that most of the tested reference-free indexes are useful indicators for tracking some of the real degradation of the calibrated or even uncalibrated imagery, but they are far from perfect to match an error or similarity measure if the clean or reference data is available.